How Will AI Transform U.S. Banks?

Prashant Mehrotra, Chief AI Officer at U.S. Bank, has a framework for what his job actually is. "I look at my role in terms of four E's," he told the Wall Street Journal in a conversation with Amit Anand, a Principal at Deloitte Consulting. "Evangelize, educate, enable, and execute on AI efforts across the enterprise."

The sequence places adoption and capability building before deployment. At a bank the size of U.S. Bank, rollout depends on alignment across business and technology teams, workforce training, and the systems and governance required to support AI use.

These steps come before execution and determine whether deployment can scale. Mehrotra presents execution as the final stage, dependent on the preceding steps.

Enterprise AI adoption has largely moved through two phases, experimentation and proof of concept, without successfully completing the transition to scaled production. Mehrotra's interview with the Wall Street Journal addresses why, and how to fix it.

"It's important to build scaling into designs from the start," he said. "The concept of reuse can be particularly valuable. By building in modular fashion with reuse in mind, it's possible to enable faster scaling throughout the enterprise."

The logic is that a platform approach to reusability allows a capability to be built once and then deployed across multiple use cases once its value is demonstrated. "You can do less building and more assembling," he said, "and the value can become exponential."

The common failure mode he identifies is organizations that treat each AI use case as a standalone build, creating redundant work, slowing deployment, and preventing the kind of compounding returns that distinguish successful AI programmes from persistent pilot cycles.

"It's less about when we as leaders think it's time and more about when users are ready," he said. The signal that an organisation is ready to move from targeted deployment to broader transformation is when employees are asking for more, not when executives are mandating it. "When people are empowered, they'll likely surprise you with their hunger for more. You know it's time to scale when your users are pulling you ahead."

That framing has a governance implication. Mehrotra advocates for what he calls a compliant-by-design approach, building compliance and risk management into AI systems from the start rather than retrofitting them after deployment.

In a regulated financial institution, this distinction is the difference between AI that can be scaled and AI that gets stopped by risk functions every time it reaches a new boundary.

Mehrotra also addressed the build-versus-buy decision directly, arguing that the pace of AI development has made a clear decision framework essential.

"With AI, the pace of change is so fast that lacking a clear decision framework and structure could lead to waste," he said.

Build for use cases that depend heavily on internal data, proprietary processes, or institutional knowledge. For everything else, evaluate whether a third-party SaaS solution will deliver the needed functionality cost-effectively, and whether what you might build internally will still be competitively differentiated by the time it is deployed.

His closing advice to other AI leaders distils a longer argument into a single imperative. Don't wait for the perfect use case. Start with something manageable. Build on it. "If not now, when? And if not me, who? The time is now."